Communities in Italian corporate networks

Transcription

1 Ownership and board interconnections 20 January 2012

2 Agenda 1 Background Listed companies 2 Ownership interconnections Board interconnections 3 Why network analysis is needed Building ownership & board networks 4 5

3 Agenda Background Listed companies 1 Background Listed companies 2 Ownership interconnections Board interconnections 3 Why network analysis is needed Building ownership & board networks 4 5

4 The modern corporation Listed companies In the modern corporation there is a clear distinction in roles between: shareholders who own the company the Board of Directors (BoD), which oversees the operations of a company on a regular basis and is elected by shareholders managers who decide on a day to day basis and report to the BoD This structure ensures managerial continuity even when the shareholder base changes

5 Listed companies Listed companies The separation between rm's management and ownership base allows the existence of stock markets, Stock markets are places where shares (the atoms which build shareholder's equity) are bought and sold among investors anonymously Each seller sets a minimum price for his share to be sold Each buyer sets a maximum price for his share to be bought When the sell-side and buy-side price ranges overlap, a trade occurs About 3 billion USD are traded, on the New York Stock Exchange only, every day

6 Ownership structure Listed companies Managers of listed companies have no idea about who are most of their shareholders However some shareholders are bigger than others and, in particular, the ownership structure of a listed company may be one of the following: Widely held: all shareholders are small Public rms with blockholders: some shareholder owns a substantial fraction of rm's equity Public rms with a controlling shareholder: a shareholder (or group of shareholders) has enough power to dominate the shareholder assembly (>50% is sucient but lower stakes may be sucient)

7 Example Shareholders in Pirelli & C. Listed companies

8 Example Shareholders in Mediobanca Listed companies

9 Example Shareholders in RCS Listed companies

10 Agenda Background Ownership interconnections Board interconnections 1 Background Listed companies 2 Ownership interconnections Board interconnections 3 Why network analysis is needed Building ownership & board networks 4 5

11 Ownership links Ownership interconnections Board interconnections As shown above, often shareholders in listed companies are other listed companies listed companies are not mutually independent analyzing companies as stand-alone entities can be misleading need for a more systemic perspective This cannot be explained by diversication (shareholders can replicate it and won't give it any additional value) Two fundamental reasons: keeping unitary control (business groups) obtaining strategic objectives (e.g. reinforcing industrial alliances)

12 Business groups The Pirelli group Ownership interconnections Board interconnections

13 Strategic alliances Ownership interconnections Board interconnections Sometimes ownership ties are due to the willingness of reinforcing strategic alliances Strategic alliances: mutual benets if cooperation incentive for unilateral deviation (free-riding) Mutual shareholdings make cooperation more likely because players internalize part of the externalizes due to cooperation which, in the absence of complete contracts would cause free-riding

14 Strategic alliances Foreign investments in China Ownership interconnections Board interconnections Well-know examples of how strategic alliances can translate in equity links are the early attempts of western corporations to enter the Chinese Telecommunication market by reinforcing their partnership with local players with ownership ties: Vodafone in China Mobile (3.2%) Telefònica in China Unom (8.8%) A similar bridge-building strategy has been followed in the oil sector (BP, Exxon, Shell), raw materials (Alcoa) and Banks (Morgan Stanley)

15 Types of strategic alliances Ownership interconnections Board interconnections Strategic alliances should be considered in a very broad tense: Not only Industrial alliances (e.g. developing a new tech) Commercial alliances (e.g. exploring new markets) but also... lobbying mutual assistance

16 Agenda Background Ownership interconnections Board interconnections 1 Background Listed companies 2 Ownership interconnections Board interconnections 3 Why network analysis is needed Building ownership & board networks 4 5

17 Directors Background Ownership interconnections Board interconnections The BoD is elected by shareholders to run the company according their interest Directors sitting in the Board are ultimately responsible of the decision making process within the rm Directors can be Executive Non-executive non-independent independent

18 Busy directors Ownership interconnections Board interconnections Some directors often serve in several boards: why? several companies may have the need to have reputed professionals in their board and the number of reputed professionals available is limited companies which are controlled by the same shareholders are likely to share part of their board a company may want one of its executives to sit in another company's board to straighten a strategic alliance, pretty much like with ownership links (here monitoring is involved, rather than an incentive)

19 Board interlocks Ownership interconnections Board interconnections If one director sits in two boards this creates a link between the two companies This link is known as board interlock There is a vast literature on board interlocks and the way they can shape company's decision process creating conicts on interests (transaction with related parties) reducing the independence of directors by creating mutual hostages

20 The BoDs in the Pirelli Group Common directors Ownership interconnections Board interconnections Director Camn Pirelli & C Pirelli RE Malacalza Vittorio X X Parazzini Enrico X X Pirelli Alberto X X Tronchetti Provera Marco X X X

21 Agenda Background Why network analysis is needed Building ownership & board networks 1 Background Listed companies 2 Ownership interconnections Board interconnections 3 Why network analysis is needed Building ownership & board networks 4 5

22 Importance of corporate networks Why network analysis is needed Building ownership & board networks Board and ownership links between rms are non-random They are the outcome of a complex interplay among companies The result is a tangled network of links which can be studied by using tools for complex network analysis It is well-known by practitioners that links between companies are important in understanding their evolution and that of the markets as a whole An example may help in understanding the concept

23 Importance of corporate networks The succession in Generali Why network analysis is needed Building ownership & board networks The role of Mediobanca (one of Italy's most inuential investment banks) in the succession battle at Assicurazioni Generali in the rst half of 2010 is describe by the Financial Times as follows: [...] it will be within the walls of Mediobanca, the Milan-based investment bank, that the future leadership of Generali will most likely be decided. Tucked away behind La Scala opera house, Mediobanca [...] retains one last vestige of its former status in the shape of its 15 per cent stake in Trieste-based Generali. In the run-up to Generali's April shareholders' meeting at which the insurers board is up for renewal, [...] Mediobanca will momentarily resume its place in the limelight as the leadership of one of Italy's great names comes up for grabs (Financial Times, February 3, 2010, French connection could decide Generali's succession, by Paul Betts).

24 The need for network analysis Why network analysis is needed Building ownership & board networks The notion of inter-company inuence and the existence of inter-company links has however, until recently, been totally neglected in academic research: why? Building a clear, objective, measure for this construct proves to be pretty dicult Network analysis may prove extremely useful in providing a quantitative measure to such an important, yet slippery, notion

25 Notation Background Why network analysis is needed Building ownership & board networks For an graph of N nodes L links we dene: A connectivity matrix A = [a i,j ] where a i,j = 1 if a link between i and j exists A weight matrix W = [w i,j ] where w i,j > 0 if a i,j = 1 and w i,j = 0 otherwise

26 Some more notation Why network analysis is needed Building ownership & board networks For undirected networks we dene: A node degree k i = j a ij = i a ij A node strength w i = j w ij = i w ij The total network weight w = 1 2 ij w ij

27 Agenda Background Why network analysis is needed Building ownership & board networks 1 Background Listed companies 2 Ownership interconnections Board interconnections 3 Why network analysis is needed Building ownership & board networks 4 5

28 Directed ownership networks Why network analysis is needed Building ownership & board networks In ownership networks, links are driven by mutual shareholdings A directed network of listed companies can be built as follows: each company is assigned a node a link between company i and company j exists in case the former has a shareholdings in the latter (a ij = 1) w i,j [0, 1) is proportional to the size of the stake

29 Removing directionality Why network analysis is needed Building ownership & board networks Often an undirected version of the ownership network is used The rationale is that what is most import an is the existence of interdependences among companies rather than its directionality If company A owns a stake in company B this creates a link between A and B but also a link between B and A An investor C linked with B should also be, indirectly, linked with A (e.g. if you buy a 5% stake in Pirelli R.E. this will link you with the whole Pirelli Group, not with Pirelli R.E. only) To remove directionality we build a network such that: { } a ij = max a ij, a ji ( ) w ij = 1 w 2 ij + w ji

30 Bipartite board networks Why network analysis is needed Building ownership & board networks Board interlocks can be used to create a undirected, unweighted bipartite network N A boards (i.e. companies) N B directors A link between director k and a board h (i.e. a ) is present if hk the director serves in the board

31 Projecting the board network Why network analysis is needed Building ownership & board networks Projections of the network can be done by creating a network of boards linked by (one or more) interlocks The projected network will have N A nodes and links will be such that: { } a ij = sup a ki a kj k A weight can be attached to the link as follows: w i,j = 2n ij n i +n j where n i = k a ki is the number of directors in board i and n ij = k a ki a kj is the number of directors in common between boards i and j

32 Agenda Background Why network analysis is needed Building ownership & board networks 1 Background Listed companies 2 Ownership interconnections Board interconnections 3 Why network analysis is needed Building ownership & board networks 4 5

33 Communities Why network analysis is needed Building ownership & board networks A community is a sub-network with many internal connection and few connections with the rest of the network Given a sub-network C this wedge in the density of links can be measured as: [ ] wij w i w j 2w i,j C (2w) 2

34 Modularity Why network analysis is needed Building ownership & board networks Consider now a partition of the network in q communities {C 1,..., C q }; the extent to which the partition has strong within-community links with respect to between-communities link is given by it modularity, dened as: Q = q h=1 i,j C h ] w i w j 2w (2w) 2 [ wij The higher Q the higher is the community structure of a partition

35 Why network analysis is needed Building ownership & board networks aims at nding the partition of a network which most correctly identies existing communities This translates in maximizing Q over all possible partitions of the network is a graph clustering algorithm in which the number q of clusters is not a priori dened The number of partitions of a N-node network grows faster than e N : exhaustive optimization of Q is impossible Several eective sub-optimal methods are available, capable of analyzing networks with up to few millions nodes Modularity optimization is not the only way to identify communities but is one of the most eective

36 Is modularity high enough? Why network analysis is needed Building ownership & board networks A high value of Q max means that a network exhibits community structure But how high has Q max to be to say that the structure is actually statistically signicant? Suppose one generates M random networks with the same degree sequence of the original network For each replication a modularity Qmax m is obtains; let µ = 1 M m Qm max and σ2 = 1 M 1 m (Qm max µ) 2 Then the null hypothesis that the original network is indeed random can be tested by z = (Qmax µ) σ

37 Communities in corporate networks Why network analysis is needed Building ownership & board networks If we apply community analysis to a corporate network, we can identify how rms are linked by the observable and unobservable forces which create ownership and board links We are not interested in communities in ownership or board networks per se but we study the community structure of these networks because they reect a deeper strategic agglomeration Paradoxically, the more obvious is the partition (i.e. the more it reects the a priori classication of an informed observer), the more the partition can be considered valid

38 Agenda Background 1 Background Listed companies 2 Ownership interconnections Board interconnections 3 Why network analysis is needed Building ownership & board networks 4 5

39 All companies listed on the Italian Stock Exchange as of January 2009 Ownership stakes come form Consob lings all stakes>2% have to be declared to Consob only voting shares are considered voting shares held by controlled companies are consolidated We have to partially de-consolidate shareholdings to avoid double counting Network is undirected

40 Example RCS Consob ling

41 The Italian ownership network At a glance

42 The Italian ownership network Characteristics

43 The Italian 15 communities in the giant component Strong community structure Q = 0.82 and z = 29.2

44 Example Pirelli Background

45 Example Mediobanca Background

46 Agenda Background 1 Background Listed companies 2 Ownership interconnections Board interconnections 3 Why network analysis is needed Building ownership & board networks 4 5

47 The Italian Board Network All directors at end 2008 Data from Consob lings Weighed undirected projection of the bipartite network

48 The Italian Board Network At a glance

49 The Italian Board Network Characteristics

50 Modularity 12 communities in the giant component Strong community structure Q = 0.59 and z = 3.82

51 Example Pirelli Background

52 Agenda Background 1 Background Listed companies 2 Ownership interconnections Board interconnections 3 Why network analysis is needed Building ownership & board networks 4 5

53 Background The community structure of the two networks is signicant but is it capturing what we want to? We test the validity of the partition in communities using two criteria: Consistency with a priori partitions Overlap

54 Is this obvious enough? To test the validity of the community structure we rst compare it with our a priori We identify 13 business groups using the accounting standard IAS 27 no business group is split between communities in the ownership network only 2 out of 13 business groups are split between communities in the board network

55 Overlap Background The underlying forces which shape both networks should be the same Accordingly the two community structures should be broadly consistent We measure the consistency of the two communities using dierent criteria (Rand Index, Adjusted Rand Index, van Dongen distance, variation of information) All these conrm that the two community structures are signicantly overlapped

56 Agenda Background 1 Background Listed companies 2 Ownership interconnections Board interconnections 3 Why network analysis is needed Building ownership & board networks 4 5

57 Listed companies are not isolated Systemic view is sometimes required Board and s capture our naive idea of interconnection Which consequence? Some empirical hypotheses can be tested once naive ideas are translated into quantitative measures Are more important companies more expensive (e.g. higher P/E)? Do marginal shareholders pay for centrality? Are stock returns of companies in the same communities abnormally correlated? Do communities provide last-resort help in troubled times?